Foundlings
A calm companion that lives inside Claude Desktop, and grounds you in the day you have actually had.
The question I was chasing
Could a calm, genuinely alive companion live inside the tool I already talk to all day?
Every daily-wisdom app hands you a generic quote with no idea what your day is like. I wanted the opposite: something that sits in the conversation where I am already working, knows the kind of day I have actually had, and offers a few grounding words for that moment rather than a fortune cookie. The real question was whether it could live right inside Claude Desktop, with no separate app and no terminal, and still feel present rather than printed.
Why it exists
Every other daily-wisdom app hands you a generic quote with no idea what your day is like. What makes context-aware reflection possible is being in the conversation where the work already happens, so instead of "here is today's stoic quote" it can say "you have been wrestling this for two hours, that is persistence, not failure."
It began as a self-care pet you fed by talking to Claude through the day. The feeding loop turned out to be the wrong heart. What the idea actually wanted was presence, not another streak to protect.
The constraints
Three hard rules shaped it. It had to install with one double-click and no terminal, so a non-technical person could have it in Claude Desktop in seconds. It had to be private: the extension carries no brain and no network of its own, so your conversation never leaves Claude. And it had to feel alive at the size it actually renders, breathing and glancing on its own, not a frozen loop that plays once and stops. Those three pulled in different directions, and most of the build was holding all three at once.
The decisions that mattered
The first attempt animated the creature in Lottie, and it died on a simple rule: you cannot tune motion you cannot watch. Keyframing a pre-made rig gave stiff, puppet-like movement, and I was judging it from stills instead of playback. So I redrew the creature in code as an addressable SVG rig, animated it with GSAP, and built a gallery that renders every pose down to twenty-two pixels so I can watch it read at the size it ships. Never author blind became the standing rule.
The bigger call was moving the whole thing inside Claude. The original plan shipped a separate menu-bar app. Asking whether it could instead live inside Claude Desktop, as a single double-click extension, turned out to be roughly ninety percent code reuse, because the creature was already a brain of pure functions and a face of pure SVG, split along exactly that seam.
The last decision was the reframe. The moment a state is triggered by Claude reading the real conversation rather than a stats engine, a richer range of expression becomes attunement, not a digital pet creeping back in. That freed the creature to mirror your day and stay alive at rest without ever asking to be fed.
What it is
A single double-click extension for Claude Desktop. Say "ground me" or type /ground, and the creature appears inline in the chat doing what you have been doing, with a short reflection written fresh for your moment fading in beneath it before it settles into a living breath. There are no stats, no streaks and nothing to feed. It runs entirely on your machine, with no brain or network of its own: Claude, already there in your conversation, chooses how it shows up and writes the words, and the creature performs.
Built with: a hand-rolled MCP server, an MCP App panel, GSAP on an addressable SVG rig, Claude, Claude Code
Where it landed
It ships as Foundlings, real and working today, installed the hand-built way rather than as a polished store listing, and that is part of the charm. Getting there meant pioneering a surface Anthropic has not documented for this use: rendering an interactive panel inside a local extension, where a too-large build fails to install silently, with no dialog and no error, so minifying the animation runtime turned out to be load-bearing rather than cosmetic. If I took it further, the open questions are the ones worth getting right slowly: how proactive it should be, whether it remembers, and whether the blob grows up into a calmer creature of its own.
Part of the Rolling Waves work archive.